KR102639267B1 - Method and apparatus for diagnosing progressive supranuclear palsy using deep learning - Google Patents

Abstract

Embodiments of the present specification are techniques for determining progressive supranuclear palsy using deep learning. Specifically, it is a technology that divides the substantia nigra region of the midbrain using a convolutional neural network and determines progressive supranuclear palsy through analysis of the volume, surface area, and shape of the divided substantia nigra region. A computer device according to an embodiment of the present specification for achieving the above-described problem includes: a memory capable of storing data and at least one module; and acquiring a plurality of tomographic images of the brain, using a network model to estimate the substantia nigra region in each of the plurality of tomographic images to identify a plurality of substantia nigra segmented regions, and substantia nigra based on the plurality of substantia nigra segmented areas. It is characterized by including a processor that controls to estimate the shape and determine whether progressive supranuclear palsy syndrome is present based on the substantia nigra three-dimensional shape.

Description

Method and apparatus for diagnosing progressive supranuclear palsy using deep learning}

Embodiments of the present specification are techniques for determining progressive supranuclear palsy using deep learning. Specifically, it is a technology that divides the substantia nigra region of the midbrain using a convolutional neural network and determines progressive supranuclear palsy through analysis of the volume, surface area, and shape of the divided substantia nigra region.

Atypical Parkinsonism refers to a degenerative disease that shows symptoms similar to Parkinson's disease but also shows additional symptoms. These atypical Parkinson's syndromes include progressive supranuclear palsy, multisystem atrophy, corticobasal nuclear degeneration, and Lewy body dementia.

Among them, progressive supranuclear palsy (PSP) is one of the representative atypical Parkinson's syndromes. It differs from Parkinson's disease in that it presents with postural instability that makes it difficult to maintain balance from the beginning of the disease. Additionally, patients with progressive supranuclear paralysis have clinical symptoms such as spasticity of the muscles around the neck and other core muscles, walking with the neck tilted back, and problems with looking downward due to impaired ability to control eye movements. Symptoms appear.

Since there are differences in treatment methods and drugs for atypical Parkinson's syndrome between healthy controls and Parkinson's disease patients, the clinical need for effective treatment of patients with respect to differential diagnosis of atypical Parkinson's syndrome is increasing. there is. Conventionally, the type and severity of Parkinson's disease and atypical Parkinson's syndrome were mainly diagnosed based on the patient's clinical symptoms, but clinical symptoms overlap even among patients, making clear differentiation of atypical Parkinson's syndrome realistically difficult.

With the recent development of Magnetic Resonance Imaging (MRI), diagnosis and follow-up of Parkinson's disease and atypical Parkinson's syndrome are being performed based on changes in MRI signals within the substantia nigra, and identifying Hummingbird signs or Morning glory signs in the patient's brain. Diagnosis is performed through qualitative analysis.

Judgment based on changes in clinical symptoms and MRI signals within the substantia nigra has the problem that analysis results may differ depending on the capabilities of the medical staff, so methods and devices to quantitatively distinguish atypical Parkinson's syndrome are problematic.

The above-mentioned background technology is technical information that the inventor possessed for deriving the present invention or acquired in the process of deriving the present invention, and cannot necessarily be said to be technology disclosed to the general public before filing the application for the present invention.

KR 10-2020-0029313 A KR 10-2019-0105452 A

Various embodiments described in this specification are proposed to solve the above-mentioned problems, and embodiments of this specification provide a method and device for determining progressive supranuclear palsy using deep learning.

A method performed by a computer device according to an embodiment of the present specification for achieving the above-described task includes acquiring a plurality of tomographic images of the brain; Using a network model, estimating a substantia nigra region in each of the plurality of tomographic images and confirming a plurality of substantia nigra segmented regions; estimating a three-dimensional shape of the substantia nigra based on the plurality of substantia nigra segmented regions; and determining whether progressive supranuclear palsy syndrome exists based on the substantia nigra three-dimensional shape.

A computer device according to an embodiment of the present specification for achieving the above-described problem includes: a memory capable of storing data and at least one module; and acquiring a plurality of tomographic images of the brain, using a network model to estimate the substantia nigra region in each of the plurality of tomographic images to identify a plurality of substantia nigra segmented regions, and substantia nigra based on the plurality of substantia nigra segmented areas. It is characterized by including a processor that controls to estimate the shape and determine whether progressive supranuclear palsy syndrome is present based on the substantia nigra three-dimensional shape.

The first convolution block according to an embodiment of the present specification for achieving the above-described task includes a first convolution layer, a first batch normalization layer, and a first activation layer ( Features include a first activation layer, a dropout layer, a second convolution layer, a second batch normalization layer, and a second activation layer. Do it as

According to one embodiment of the present invention, the substantia nigra region of the midbrain can be segmented in a magnetic resonance image using a nested U-Net neural network.

According to one embodiment of the present invention, progressive supranuclear palsy can be objectively determined by quantitatively analyzing the volume, surface area, and shape of the substantia nigra.

The effects of the present invention are not limited to the effects mentioned above.

Figure 1 is a diagram schematically showing a process for estimating the three-dimensional shape of the substantia nigra by a diagnostic device according to an embodiment of the present invention.
Figure 2 is a diagram showing the structure of a network model according to an embodiment of the present invention.
FIG. 3A is a diagram showing the structure of a first convolution block according to an embodiment of the present invention.
FIG. 3B is a diagram showing the structure of a second convolution block according to an embodiment of the present invention.
Figure 4 is a flowchart showing the operation of determining whether a device has progressive supranuclear palsy syndrome according to an embodiment of the present invention.
Figure 5 is a flowchart showing an operation in which a device performs network learning according to an embodiment of the present invention.
Figure 6 is a block diagram illustrating the configuration of a diagnostic device according to an embodiment of the present invention.

Since the present invention can be variously modified and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. The effects and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various forms.

Each block may represent a module, segment, or portion of code containing one or more executable instructions to execute a specific logical function. It should be noted that in other embodiments, it is possible for the functions mentioned for each block to be executed differently from the order described. For example, even if two blocks are shown one after another, the functions described for each block may be performed substantially simultaneously, or may be performed in reverse order as execution conditions or environments vary. In the following examples, singular terms include plural terms unless the context clearly dictates otherwise.

In the following embodiments, terms such as include or have mean that the features or components described in this specification exist, and do not exclude the possibility of adding one or more other features or components.

Instructions executed through a processor of a computer or data processing equipment may create a means of performing each function described with reference to a flowchart or block diagram. Instructions may be mounted on a computer or data processing equipment to create processes that are executed on the computer or the like to perform a series of operation steps.

At this time, the term '~ part' used in this embodiment means a component that performs a specific function performed by software or hardware such as FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit). However, '~ part' is not limited to being performed by software or hardware. The '~ part' may exist in the form of data stored in an addressable storage medium, and one or more processors may be configured to execute a specific function.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. Software and/or data may be used on any type of machine, component, physical device, virtual equipment, computer storage medium or device to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

An 'image' according to the present invention may be a still image or an image of one frame extracted from a video composed of a plurality of consecutive frames. Additionally, the image according to the present invention may be either an inference target image that is the target of inference or a learning image that is the target of learning.

The learning image according to the present invention is a training data set for learning a network model, for example, images for image segmentation are used as features, and images with labeled segmentation areas are used as labels. It can refer to a data set of . Features according to an embodiment of the present invention may be a plurality of tomographic images of the brain, for example, an R2 Star map or susceptibility-weighted imaging (SWI), and the labels may be the substantia nigra segmented area. It can be.

The neural network model or network model according to the present invention is a representative example of an artificial neural network model that simulates brain nerves, and is not limited to, for example, an artificial neural network model using a specific algorithm.

The network model according to the present invention may be U-Net or Nested U-Net, and the network model is a convolutional neural network for image segmentation and can simultaneously perform classification and localization.

Figure 1 is a diagram schematically showing a process for estimating the three-dimensional shape of the substantia nigra by a diagnostic device according to an embodiment of the present invention.

The diagnostic device according to the present invention can acquire an R2 Star map 11 and/or a susceptibility-weighted image 13, which are three-dimensional images of the entire brain. Afterwards, the diagnostic device can extract the image 10 showing the substantia nigra from the magnetic resonance image and use it as input to the network model. For example, the diagnostic device may extract an image 10 showing the substantia nigra from the R2 Star map 11 and/or the susceptibility-emphasized image 13 and input it into the network model 20. The image 10 showing the substantia nigra may refer to an image including the substantia nigra region among three-dimensional images of the entire brain.

In addition, the diagnostic device uses both the R2 Star map (11) and the susceptibility-weighted image (13) taken of the same target brain as input to the network model (20), or one of the R2 Star map (11) and the susceptibility-weighted image (13). can be used as an input to the network model (20).

The diagnostic device inputs the image 10 into the network model 20 and outputs first result data 23 and second result data 21, and the first result data 23 and the second result data 21 Based on the average of , substantia nigra segmented area data 25 can be generated by selecting an area greater than a threshold. The substantia nigra segmented area data 25 refers to data obtained by segmenting the substantia nigra area from the image 10 in which the substantia nigra area is revealed.

Since the image containing the substantia nigra corresponds to the segmentation data 25 of the substantia nigra, which is the model output of the network, the diagnostic device can segment the substantia nigra for all images containing the substantia nigra. ), the diagnostic device can generate data 31 by segmenting a plurality of substantia nigra regions.

The diagnostic device can estimate the substantia nigra three-dimensional shape 33 using data 31 obtained by segmenting a plurality of substantia nigra regions. The diagnostic device may obtain determined substantia nigra information regarding the volume, surface area, and curvature of the substantia nigra three-dimensional shape 33 and perform diagnosis based on the volume, surface area, and curvature of the substantia nigra three-dimensional shape 33.

Figure 2 is a diagram showing the structure of a network model according to an embodiment of the present invention. The network model 20 corresponds to the network model 20 in FIG. 1. The network model according to the present invention may be U-Net, Nested U-Net, or a modified form of U-Net, and the network model is a convolutional neural network for image segmentation, classification and localization ( Localization) can be performed simultaneously.

Referring to Figure 2, for explanation purposes, the structure of the network model 20 is assumed to be a modified version of the Nested U-Net. The structure of the network model 20 is composed of nodes (X ^i,j ) composed of the first convolution block, down-sampling, up-sampling, and skip path. You can. At this time, i is the down sampling index and j is the skip path index.

At ^node

The operation of the first convolution block can be performed at the node (X ^i,j ). As an example, the operation of the first convolution block may correspond to the operation in FIG. 3A. The first convolution block includes a first convolution layer, a first batch normalization layer, a first activation layer, a dropout layer, and a second convolution layer. It may include a second convolution layer, a second batch normalization layer, and a second activation layer.

Down-Sampling can receive a feature map as input, perform max pooling, and deliver it to a lower level node related to the backbone. According to an embodiment of the present invention, the size of the image can be halved by performing 2Х2 max pooling and stride 2. Afterwards, the depth of the channel can be doubled by doubling the filter value of the first convolution block. For example, assuming that the output feature map size of X ^0,0 is 256Х256Х32, ^the output feature map size of

Up-Sampling can receive a feature map as input, perform deconvolution, and transmit it to a higher-level node.

In the skip pathway, an operation is performed to concatenate the results of a node located at the same level with the results upsampled from a lower level node, and a value can be transmitted to the node. As explained with reference to Figure 2, node ^{X 1,3} uploads i) the feature maps output from nodes ^{X 1,0 ,} The value connected to the sampling ( ^Up _- Sampling) value

This can be expressed by Equation 1 below.

[Equation 1]

In Equation 1 ^{, x i ,j} is the output value of the j is the skip path index.

In the network model 20 according to the present invention ^, the basic final node ²⁰¹ can be defined as At this time, J means the maximum value of the skip path index. Referring to FIG. 2, J = 5, the primary final node 201 may be X ^0,5 , and the secondary final node 203 may be X ^0,4 .

In the network model 20 according to the present invention, an operation may be performed using output data from the basic final node 201 as input to the second convolution block. Accordingly, first result data may be generated.

In the network model 20, an operation may be performed using output data from the auxiliary final node 203 as input to the second convolution block. Accordingly, second result data may be generated.

Final result data may be generated by averaging the first result data and the second result data and selecting an area greater than a threshold. According to an embodiment of the present invention, the final result data may be data 25 obtained by segmenting the substantia nigra region from the image 10 in which the substantia nigra region is revealed.

FIG. 3A is a diagram showing the structure of a first convolution block according to an embodiment of the present invention.

The first convolution block 300a may be a block for performing operations at nodes X ^i,j . Additionally, the first convolution block 300a may be a block for performing multiple convolution layer operations.

The first convolution block 300a according to an embodiment of the present invention includes a first convolution layer 301a, a first batch normalization layer 303a, and a first activation layer ( first activation layer (305a), dropout layer (307a), second convolution layer (308a), second batch normalization layer (309a), second It may include a second activation layer 311a.

The first convolution block 300a performs calculations based on the feature map and filter received as input to the node. In the first convolution layer 301a and the second convolution layer 308a, features of the input feature map are extracted using a filter. The first convolution layer 301a and the second convolution layer 308a output a new feature map based on the filter. The distribution of data may be normalized in the first batch normalization layer 303a and the second batch normalization layer 309a. In the first activation layer 305a and the second activation layer 311a, the weighted sum of input data may be converted into output data. The activation layer may use at least one of, for example, ReLU, elu, Maxout, or tanh. In the dropout layer 307a, an operation to prevent overfitting by removing random neuron nodes may be performed.

The first convolution block 300a performs a first convolutional layer 301a operation, a first batch normalization layer 303a operation, and a first activation layer 305a based on the feature map and filter value received as input to the node. The operation, dropout layer (307a) operation, second convolution layer (308a) operation, second batch normalization layer (309a) operation, and second activation layer (311a) operation can be performed sequentially and the result can be output. .

FIG. 3B is a diagram showing the structure of a second convolution block according to an embodiment of the present invention.

In the network model 20 according to the present invention, an operation may be performed using output data from the basic final node 201 as input to the second convolution block. Accordingly, first result data may be generated. In the network model 20, an operation may be performed using output data from the auxiliary final node 203 as input to the second convolution block. Accordingly, second result data may be generated.

Referring to FIG. 3b, the second convolution block 300b is composed of a convolution layer 301b, a batch normalization layer 303b, and an activation layer 305b. It can be input, and the output data from the basic final node 201 and the output data from the secondary final node 203 can be used as input.

The second convolution block 300b may sequentially perform the convolution layer 301b operation, the batch normalization layer 303b operation, and the activation layer 305b operation and output the results.

Figure 4 is a flowchart showing the operation of determining whether a diagnostic device has progressive supranuclear palsy syndrome according to an embodiment of the present invention.

Referring to FIG. 4, the diagnostic device may pre-learn a network model in step S110. For example, a plurality of learning tomographic images of the brain and substantia nigra region labeling images corresponding to each of the plurality of learning tomography images are acquired, and the parameters of the network model representing the correlation between the learning tomography images and the substantia nigra region labeling images are adjusted to form a network. You can learn a model. As an example, the network model may correspond to the network model 20 in FIG. 2.

In step S120, the diagnostic device may generate representative substantia nigra three-dimensional shapes of progressive supranuclear palsy syndrome patients, Parkinson's disease patients, and control groups.

< Progressive supranuclear palsy syndrome >

By inputting the R2 Star map and/or susceptibility-weighted image of the brain of a patient with progressive supranuclear palsy syndrome to the pre-trained network model, the substantia nigra segmented area of the patient with progressive supranuclear palsy syndrome can be confirmed. The diagnostic device can estimate the three-dimensional shape of the substantia nigra of a patient with progressive supranuclear palsy syndrome using the substantia nigra division area.

By repeating the above process for a number of patients with progressive supranuclear palsy syndrome, representative or average values for the volume, surface area, and curvature of the substantia nigra three-dimensional shape can be determined and a representative substantia nigra three-dimensional shape can be generated. For example, representative values or average values of the volume, surface area, and curvature of the substantia nigra three-dimensional shape of a patient with progressive supranuclear palsy syndrome according to an embodiment of the present invention may be referred to as first substantia nigra information.

<Patients with Parkinson’s disease>

By inputting the R2 Star map and/or susceptibility-weighted image of the brain of a Parkinson's disease patient into the pre-trained network model, the substantia nigra segmented area of the Parkinson's disease patient can be confirmed. The diagnostic device can estimate the three-dimensional shape of the substantia nigra of a Parkinson's disease patient using the substantia nigra segmented area.

By repeating the above process for a number of Parkinson's disease patients, representative or average values for the volume, surface area, and curvature of the substantia nigra three-dimensional shape can be determined and a representative substantia nigra three-dimensional shape can be generated. For example, representative values or average values of the volume, surface area, and curvature of the three-dimensional shape of the substantia nigra of a Parkinson's disease patient according to an embodiment of the present invention may be referred to as second substantia nigra information.

<Control group>

By inputting the R2 Star map and/or susceptibility-weighted image of the brain of the control group without abnormal findings to the pre-trained network model, the substantia nigra segmented area of the control group can be confirmed. The diagnostic device can estimate the three-dimensional shape of the control group's substantia nigra using the substantia nigra segmented area.

By repeating the above process for a number of control groups, representative or average values for the volume, surface area, and curvature of the substantia nigra three-dimensional shape can be determined and a representative substantia nigra three-dimensional shape can be generated. For example, representative values or average values of the volume, surface area, and curvature of the substantia nigra three-dimensional shape of the control group according to an embodiment of the present invention may be referred to as third substantia nigra information.

The diagnostic device may acquire a plurality of tomographic images of the brain to be diagnosed in step S130. For example, an R2 Star map and/or susceptibility-weighted image of the brain that is a diagnostic target can be obtained.

In step S140, the diagnostic device may identify a plurality of substantia nigra segmented areas by estimating the substantia nigra region from each of the plurality of tomographic images using a pre-trained network model.

In step S150, the diagnostic device may estimate the three-dimensional shape of the substantia nigra based on a plurality of substantia nigra segmented regions. For example, the three-dimensional shape of the substantia nigra can be estimated by sequentially stacking multiple substantia nigra segmented regions.

In step S160, the diagnostic device may determine whether progressive supranuclear palsy syndrome is present based on the substantia nigra three-dimensional shape. For example, the substantia nigra three-dimensional shape of the brain to be diagnosed can be estimated, and determined substantia nigra information including the volume and surface area of the substantia nigra three-dimensional shape and the curvature of the substantia nigra three-dimensional shape can be obtained. The diagnostic device may compare the determined substantia nigra information with the first substantia nigra information, the second substantia nigra information, and the third substantia nigra information, and determine whether progressive supranuclear palsy syndrome is present based on the substantia nigra information having the closest value.

Figure 5 is a flowchart showing an operation in which a diagnostic device performs network learning according to an embodiment of the present invention.

Referring to FIG. 5 , the diagnostic device may acquire a plurality of learning tomography images of the brain in step S210.

In step S220, the diagnostic device may acquire a substantia nigra region labeling image corresponding to each of the plurality of learning tomographic images. At this time, the labeling image may refer to an image in which the medical staff directly segmented the substantia nigra region for each learning tomographic image.

The diagnostic device may determine a plurality of learning tomographic images of the brain and substantia nigra region labeling images corresponding to each of the plurality of learning tomographic images as a learning data set. For example, the learning data set is a data set that uses a plurality of learning tomographic images of the brain for image segmentation as features and a labeling image of the substantia nigra region corresponding to each of the plurality of learning tomographic images as labels. It can mean.

The diagnostic device may perform network model learning for substantia nigra segmentation in step S230. The network model can be learned by adjusting the parameters of the network model representing the correlation between the learning tomographic image and the substantia nigra region labeling image. As an example, the network model may correspond to the network model 20 in FIG. 2.

Figure 6 is a block diagram illustrating the configuration of a diagnostic device according to an embodiment of the present invention.

The diagnostic device 600 is shown as being comprised of a memory 601 and a processor 603, but is not necessarily limited thereto. The memory 601 and the processor 603 may each exist as one physically independent component.

The memory 601 may store various data for the overall operation of the diagnostic device 600, such as a program for processing or controlling the processor 603 in the diagnostic device 600. The memory 601 can store a number of running application programs, data for operating the diagnostic device 600, and commands. The memory 601 may be implemented as internal memory such as ROM or RAM included in the processor 603, or may be implemented as a memory separate from the processor 603.

The processor 603 may be configured to generally control the diagnostic device 600.

Specifically, the processor 603 can control the operation of the diagnostic device 600 using various programs stored in the memory 601 of the diagnostic device 600. The processor 603 may include a CPU, RAM, ROM, and a system bus. The processor 603 may be implemented as a single CPU or multiple CPUs (or DSP, SoC). In one embodiment, the processor 603 may be implemented as a digital signal processor (DSP), a microprocessor, or a time controller (TCON) that processes digital signals. However, it is not limited to this. , central processing unit (CPU), micro controller unit (MCU), micro processing unit (MPU), controller, application processor (AP), or communication processor (CP) )), may include one or more of ARM processors, or may be defined by the corresponding term. In addition, the processor 603 may be implemented as a SoC (System on Chip) with a built-in processing algorithm, LSI (large scale integration). It can also be implemented in the form of an FPGA (Field Programmable Gate Array).

For example, the processor 603 acquires a plurality of tomographic images of the brain, uses a network model to estimate a substantia nigra region in each of the plurality of tomographic images, identifies a plurality of substantia nigra segmented regions, and identifies the plurality of substantia nigra segmented regions. The substantia nigra three-dimensional shape can be estimated based on the segmented area, and control can be made to determine whether there is progressive supranuclear palsy syndrome based on the substantia nigra three-dimensional shape.

As described above, although the embodiments have been described with limited examples and drawings, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

11: R2 Star Map
13: Susceptibility highlighting video
20: Network model
201: Primary final node
203: Secondary final node
21: Second result data
23: First result data
25: Substantia nigra segment data
300a: first convolution block
300b: second convolution block
301a: first convolution layer
301b: convolution layer
303a: First batch normalization layer
303b: Computational, batch normalization layer
305a: first activation layer
305b: Computation, activation layer
307a: Dropout layer
308a: second convolution layer
309a: Second batch normalization layer
31: Data dividing multiple substantia nigra regions
305b: Activation layer
33: Substantia nigra three-dimensional shape
600: Diagnostic device
601: Memory
603: Processor

Claims (12)

In a method performed by a computer device,
Obtaining a plurality of tomographic images of the brain;
Using a network model for image segmentation, estimating a substantia nigra region in each of the plurality of tomographic images and confirming a plurality of substantia nigra segmentation regions;
generating a three-dimensional shape of the substantia nigra based on the plurality of substantia nigra segmented regions; and
A method comprising determining whether progressive supranuclear palsy syndrome exists based on the substantia nigra three-dimensional shape. According to claim 1,
The network model for image segmentation is,
A network for image segmentation that acquires a plurality of learning tomographic images of the brain and a substantia nigra segment labeling image corresponding to each of the plurality of learning tomography images, and indicates an association between the learning tomography image and the substantia nigra segment labeling image. A method of adjusting the parameters of a model to characterize it as pre-trained. According to claim 1,
The step of determining whether progressive supranuclear palsy syndrome is present based on the substantia nigra three-dimensional shape is,
Obtaining determined substantia nigra information including the volume and surface area of the substantia nigra three-dimensional shape and the curvature of the substantia nigra three-dimensional shape; and
Comparing first substantia nigra information, second substantia nigra information, and third substantia nigra information with the determined substantia nigra information,
The first substantia nigra information is information related to the three-dimensional shape of the substantia nigra of a patient with progressive supranuclear palsy syndrome estimated using the network model for image segmentation,
The second substantia nigra information is information related to the three-dimensional shape of the substantia nigra of a Parkinson's disease patient estimated using the network model for image segmentation,
The third substantia nigra information is information related to the shape of a normal person estimated using the network model for image segmentation. delete According to claim 1,
A method characterized in that the network model for image segmentation uses a U-Net or Nested U-Net model for object segmentation. In computer devices,
a memory capable of storing data and at least one module; and
Acquire a plurality of tomographic images of the brain, use a network model for image segmentation, estimate the substantia nigra region in each of the plurality of tomographic images, identify a plurality of substantia nigra segmentation regions, and segment the plurality of substantia nigra. A processor that generates a substantia nigra three-dimensional shape based on the area and controls to determine whether progressive supranuclear palsy syndrome is present based on the substantia nigra three-dimensional shape.
A computer device comprising: According to clause 6,
The network model for image segmentation is,
A network for image segmentation that acquires a plurality of learning tomographic images of the brain and a substantia nigra segment labeling image corresponding to each of the plurality of learning tomography images, and indicates an association between the learning tomography image and the substantia nigra segment labeling image. A computer device characterized by pre-training by adjusting the parameters of a model. According to clause 6,
The processor,
Obtain determined substantia nigra information including the volume, surface area, and curvature of the substantia nigra three-dimensional shape, and compare the determined substantia nigra information with first substantia nigra information, second substantia nigra information, and third substantia nigra information,
The first substantia nigra information is information related to the three-dimensional shape of the substantia nigra of a patient with progressive supranuclear palsy syndrome estimated using the network model for image segmentation, and the second substantia nigra information is information related to the three-dimensional shape of the substantia nigra estimated using the network model for image segmentation. Information related to the estimated three-dimensional shape of the substantia nigra of a Parkinson's disease patient, and the third substantia nigra information is information related to the shape of a normal person estimated using the network model for image segmentation. delete According to clause 6,
A computer device characterized in that the network model for image segmentation uses a U-Net or Nested U-Net model for object segmentation. According to clause 6,
The network model for image segmentation includes a first convolution layer, a first batch normalization layer, a first activation layer, a dropout layer, A computer device comprising a convolutional block comprising a second convolution layer, a second batch normalization layer, and a second activation layer. . According to clause 6,
The network model for image segmentation is characterized in that it is configured to generate final result data by connecting first result data and second result data,
The first result data is data obtained by performing a convolutional block operation on the output data of the primary final node, and the second result data is data obtained by performing a convolutional block operation on the output data of the auxiliary final node. Device.

Images